Scholarlink

<p>Discovering the <i>k</i> highest utility itemsets in a database without a predefined threshold is addressed by the top‑<i>k</i> high utility itemset mining (top‑<i>k</i> HUIM) problem. These itemsets help analyze customer behavior in retail. But previous algorithms do not take categorized information into consideration and assume that the profit of items is always a stable, positive value. A new type of database, more suitable for real-world scenarios, is presented as a solution to these limitations, and the challenge of extracting the top‑<i>k</i> cross-level high utility itemsets from this database is also explored. The paper presents the TKCUN algorithm as a solution to the problem. It employs strict upper limits, such as positive subtrees and positive local utility, and utilizes efficient pruning techniques, including sorting items by type and level. To further enhance the effectiveness of these upper bounds, the study proposes and applies threshold strategies that utilize the utility of single items, pairs of items, transactions involving items of the same level, and, most importantly, multi-level itemsets. Experiments show that TKCUN mines itemsets effectively while raising thresholds through multi-level itemset utilities, achieving up to 180 × faster execution and reducing memory use by approximately 50%.</p>

错误：搜索内容不能为空，请输入英文关键词
错误：关键词超出字数限制，请精简
高级检索

Efficient mining of top-K cross-level high utility itemsets on unstable profit databases

  • 期刊论文

  • 发表日期：

    2026年3月11日

查看全文
  • N. T. Tung,
  • Duc-Lung Vu,
  • Loan T. T. Nguyen

摘要

Discovering the k highest utility itemsets in a database without a predefined threshold is addressed by the top‑k high utility itemset mining (top‑k HUIM) problem. These itemsets help analyze customer behavior in retail. But previous algorithms do not take categorized information into consideration and assume that the profit of items is always a stable, positive value. A new type of database, more suitable for real-world scenarios, is presented as a solution to these limitations, and the challenge of extracting the top‑k cross-level high utility itemsets from this database is also explored. The paper presents the TKCUN algorithm as a solution to the problem. It employs strict upper limits, such as positive subtrees and positive local utility, and utilizes efficient pruning techniques, including sorting items by type and level. To further enhance the effectiveness of these upper bounds, the study proposes and applies threshold strategies that utilize the utility of single items, pairs of items, transactions involving items of the same level, and, most importantly, multi-level itemsets. Experiments show that TKCUN mines itemsets effectively while raising thresholds through multi-level itemset utilities, achieving up to 180 × faster execution and reducing memory use by approximately 50%.